Low-Rank Discriminative Broad Learning System for Hyperspectral Image Classification

The Broad Learning System (BLS) has gained extensive applications in hyperspectral image classification due to its concise network architecture and efficient computational performance. However, the existing traditional architecture of the broad learning method has obvious limitations in the feature extraction process. To overcome this limitation, this paper proposes a Low-Rank Discriminative Broad Learning System (LRDBLS). The low-rank constraint not only effectively removes noise from hyperspectral data but also extracts deeper latent information embedded in raw data, making our model highly suitable for hyperspectral image (HSI) classification tasks.